APBRmetrics

[back2newbelf]

A friend of mine wrote a brute force regularized adjusted +/- program. Basically it's changing player ratings as long as it finds improvements with the errors in the observations, while accounting for the penalty factor which is part of regularized adjusted +/-.
All observations were randomly split in three groups to find the best penalty value through crossvalidation.
I can't say how good this approximation really is. I do know that there is a 0.999 correlation between the results from different starting points. I guess that's a good thing.
Data is from the start of the 2005-2006 season to the 20th of November 2011. All years are weighted equally. This is probably far from perfect if you wanted to predict future games

Offense and defense is per 100 possessions, sum is per 200 possessions

http://www.docdroid.net/4t7/offense.xls.html
http://www.docdroid.net/4t8/defense.xls.html
http://www.docdroid.net/4t6/sum.xls.html

[Ilardi]

Interesting work. Did your friend happen to pass along the actual lambda value(s) his model settled on? And did he include playoff as well as regular season observations?

[back2newbelf]

Regular season only

Lamdba was 22500. I think Joe Sill had similar rating ranges with a lamdba of 3000/2000. Not sure where the difference comes from

[back2newbelf]

Now with Coaching!

Each teams' coach gets treated as a 6th man on the court.
Errors on the test sets did not hugely improve through this, but they didn't degrade either (Improvement was ~0.05%).
It seems that larger penalty (lambda) values for the coaches, compared to those for the players, help a bit.
Lambda was 25000 for the players, 45000 for the coaches

Players and coaches (Coaches have the prefix 'Coach_')
http://www.docdroid.net/4yy/defense.xls.html
http://www.docdroid.net/4yz/offense.xls.html
http://www.docdroid.net/4z0/sum.xls.html

Coaches by themselves
http://www.docdroid.net/4yv/coach-defense.xls.html
http://www.docdroid.net/4yw/coach-offense.xls.html
http://www.docdroid.net/4yx/coach-sum.xls.html

Scott Skiles and the VanGundy brothers shine

Rating changes
http://www.docdroid.net/4yu/change.xls.html

Players with highest positive change in their ratings through addition of coaches: Augustin, D.J.(+1.35); O'Neal, Shaquille; Rose, Derrick; Diaw, Boris; Battie, Tony; Kidd, Jason; Milicic, Darko; McGuire, Dominic; Jianlian, Yi; Mason, Desmond(+0.65)

and highest negative change:
Bledsoe, Eric(-0.7); Bogans, Keith; Daniels, Marquis; Erden, Semih; Deng, Luol; Hinrich, Kirk; Jennings, Brandon; Westbrook, Russell; Lewis, Rashard; Mbah a Moute, Luc(-1.35)

[DSMok1]

Okay--that's really cool. I had never thought of considering coaches as the 6th man!

EDIT: can you post the standard errors for the estimates or, alternatively, the number of possessions each player had (thus the "weighting"?)
_________________
GodismyJudgeOK.com/DStats
Twitter.com/DSMok1

[Crow]

Thanks for doing this and sharing it. I had wondered if including the coach as the "6th man" could work out. Look forward to reviewing the data.

it might be worth noting the coaching values are for the 5+ season time=period and are not for their entire career. I had to catch myself and remember that when looking at a few names.


Brown, Scott and Kuester are the only active coaches in the bottom 20%. Carlisle, Adelman and McMillan just outside that group on this metric. Jackson estimated slightly below neutral.


There are issues or things to stay aware of with the ratings for coaches on teams which had no other coach / "substitute" during the period, only one substitute (the contrast between them might get too strong and overpower the comparison to the rest of the league of coaches) or many (some smaller, less accurate sample sizes compared to the other coaches). Are the ratings for coaches who coached 2 teams better? Maybe. Agree that seeing the estimated error term would add to the data consideration.

[back2newbelf]

Now with individual strength of schedule (and number of possessions)

http://www.docdroid.net/5ir/sos.xls.html

"offensive partners" is the average of the offensive rating of those who were on the court with this player during offensive possessions
"defensive opponents" is the average of the defensive rating of the defending players while this player was attacking
and so on..

"easiness" is (strength of your partners) "minus" (strength of your opponents)

Lots of (former) Mavericks, Spurs and Celtics at the top
Young players and rookies at the bottom

This could probably be useful to those working with (mostly offensive) BoxScore numbers


The .5 in number of possessions comes from the very old bbv files, which had possessions not yet split into offensive and defensive

[bbstats]

This is really cool. I wonder how similar your friend's method is to Excel's Evolutionary method.
_________________
http://thebasketballdistribution.blogspot.com

http://twitter.com/bbstats

[back2newbelf]

This season only:

http://www.docdroid.net/5ka/2011.xls.html
Format: Offense(100 possessions)|Defense(100 possessions)|Sum

That early in the season it is obviously of limited use and will produce funny results

So far we have:

offensive player of the year: Hedo Turkoglu (he is also the worst defender though)
defensive player of the year: Darrell Arthur

While it kind of agrees with the media on the MVP race, Nowitzki/Garnett/Ginobli/super-friends all look good, it couldn't disagree more on the Rookie-of-the-year-race, putting Jeff Adrian, Landry Fields and Evan Turner at the top. John Wall is supposed to be 9th worst of all players, Griffin 7th worst
From looking at the top rated players one would think the best basketball age is 35

Also, Shane Battier is suddenly listed as a horrible defender and Chuck Hayes, who used to rock this rating, is the 10th worst player in the league

[DSMok1]

[mathayus]

[back2newbelf]

[back2newbelf]

I think we have the error fixed that made lambda so huge. At least it looks more sane now, being around ~2500 for a single season.

Single season approximated RAPM now gets published on http://stats-for-the-nba.appspot.com/ and probably updated every two weeks or so.

The site also contains data from the latest multiyear analysis which included coaches and tried different lambdas for offense and defense for both players and coaches. They were found to be: Offense: 2500, Defense: 7500, Offense(Coach): 6000, Defense(Coach): 4500.
Unfortunately the difference in error on the test sets between this method and using players only with just one lamdba is minimal.

[DSMok1]

[deepak]